Really good rundown on how AI is funded/propped up and the sheer amount of debt its generating.
AI fakes are taking up more of fact-checkers’ attention, but going viral is a poor guide to their potential harm. Full Fact’s study of 112 examples found that widely viewed fabrications could have limited impact, while smaller ones reached people able to act on them. fullfact.org/technology/fact-c…
This is what we were all sort of thinking back at my old job (in the months before I left), so it's nice to have actual data backing that sentiment up. The engineers weren't happy, but management kept forcing it on us.
Saw an article today where someone earnestly made the case that "engineers" don't need to be able to understand the code anymore, because we can just ask the LLM to describe what it does in plain language. You realise these things are still fallible right? What if its explanation is total BS? What's your fallback when the code is broken and your LLM is lying to you? It really feels like full-on psychosis. I suspect they just don't want to be developers anymore but don't want to admit it
The problem with the compulsory imposition of AI in higher education is that it forces teachers, such as myself, to spend time developing new teaching and assessment strategies to *contain* the negative impact of a product nobody asked for and that nobody needs...😑
My personal website is live: gcheseaux.ch A CV in a word processor file became a bilingual site with a blog, generated PDF CVs, accessibility checks and a small build script. Built over a few evenings with Claude Code and Mistral Vibe, and every line reviewed by me. The first blog post tells the whole story, including what I did not let AI do: gcheseaux.ch/blog/how-i-built-… #AI #ClaudeCode #WebDev #BuildInPublic
trying to imagine what it would mean for LLMs to make a pile of breakthroughs in existential phenomenology or psychoanalysis
Interesting news from work. A good friend works in a position that sees what's happening on the AI front at work. Said that leadership has brought in a consulting group that evaluates where AI can replace humans and they're currently evaluating my department of around 200 folk. Now, I've been saying this could happen, but only my buddy and I agree, maybe one other person. Sooo... RIP our jobs in 2027 probably.
An upside of this whole thing is we've stopped being so down on how we code because like, would I call someones drawing process bad? No. It's all inherently part of it being an artform.
LLMs are non-deterministic, but most business processes are deterministic. The trillion-dollar answer? If it can be deterministic, keep it deterministic. Don't pay tokens twice. #AI #LLM #Enterprise
to be fair, this is a typical experience before AI, they've simply swapped the phone firewall's people for computers
Whose idea of "better" is an AI putting into its report? In the third OrgLens note, I explore recurring organisational problems, conflicting views, and missing voices. One finding: a real quotation can still be used to support a conclusion the speaker never expressed. So checking the words isn’t enough. Also, TL;DR: The tool can show evidence; people still need to judge what it means. substrate.dougbelshaw.com/orgl… #SystemsThinking #AI
Apparently Grammarly and similar things use enough AI that they are setting off AI detectors. So I asked my students to switch that stuff off, I will take bad grammar over word salad.
TL;DR: Michael Smith has been sentenced to 18 months in prison for using bots to manipulate music-streaming platforms, falsely generating up to $8 million in royalties from his A.I.-created tracks. nytimes.com/2026/10/07/arts/mu… #law #tech #legaltech ⚖️ 🤖 #autosum
i have a headache today from the amount of brigading by anti-ai people accusing organic indie devs of using ai or being bots when they aren't. now people are saying 'we arent extreme enough' with the anti-ai approach. the witchhunting has gotten so bad we had to ban the term 'slop' in our spaces.
An important reframing by two of my favorite people. theconversation.com/ai-behaves…
this has always been the way i approached programming and ummm working inside corporations was a rude awakening as was all the framework slop (the artisanal kind)
Software designs that help mitigate or prevent Al induced deskilling are going to become more and more important in the very near future. #software #design
How do I skeet out that I think people have gotten better at dicussing AI but at a rate that will still doom us all to letting the worse parts of AI destroy our lives... No, I didn't just do it simply for the fact that people think language translation/code execution/art gen are all different ->
“Listen, learn and deliver better”: We are rightfully worried about the threats of AI, and without gov’t action the consequences could be dire. But let’s take a moment to consider how AI could improve gov’t itself, by managing complexity and finding solutions that humans struggle to see. nytimes.com/2026/10/06/opinion…
If I had a dime for every time I brought up the Therac-25 to a coworker, new hire or intern and they gave me a blank stare because nearly no computer science program has a useful ethics prerequisite, I would have a dime for virtually every person I've ever worked with professionally.
Yes AI driven game mods will be HUGE - then games will be entirely AI driven, then movies. There will be expnsive bespoke 'human made' options but they'll be the exception. Our media will mostly be GenAI shotly. This all ends terribly poorly I'm afraid.
Okay so here’s a good use for LLMs: someone write an extension that rewrites news from the West in the style Western news about other parts of the world, and vice versa.
Workers Against Slopaganda! Next week the Irish Government will host its International AI Summit in Dublin, as a counter the Digital and Techworker Alliance (DATA) will host a Pro-Worker Summit with some great speakers. datacwu.ie/ai-counter-summit/ #mastodaoine
And let me tell you, a lot of software development wasn't exactly running at peak efficiency *before* Claude ruined it for everyone.
It occurs to be that there is a contradiction about using AI and not using it because you think it is evil. If all the people who hate AI like ChatGPT wanted to strike a blow against it they would use the "free" versions constantly. Although maybe not so good for the environment in the short term it would be digging the financial hole that spells the death of these companies faster. (I may not have fully thought through this idea) #ai #uspol
I can't stop thinking about that phone interaction I had earlier and why it bothers me so much. It's the fact it tried to pass as human. And sure, technically it didn't explicitly DENY that it was AI, but the way it *evaded* the question doesn't sit right with me.
To counter AI, an illustrator in Balbriggan offers to draw posters for barter. Lauren Fernandes noticed menus and flyers all over with the same sterile look, and thought: what if she could help out small businesses which can't afford creative work, usually? dublininquirer.com/to-counter-…
Seeing how poorly US based AI companies are managed, their lack of security, ethics and respect for humanity, I realize that European-based founders have a great opportunity to develop ethical, human-centered, secure AI models.
The US risks ceding AI governance leadership to China and the EU as Trump announces voluntary AI commitments from major tech companies. The agreement lacks legal enforcement, sparking debate over whether America can maintain its edge in AI standards. scmp.com/opinion/world-opinion… #China #Tech #AIPolicy
Meanwhile, various mathematicians that I know were rushing to post results before the OpenAI release. Some of them used LLMs to come up with ideas to varying degrees. The rushed process is far from ideal, and those who still care about craft will be making a lot of further edits.
(For the uninitiated, the latter acronym is reinforcement learning from human feedback.) In other words, LLMs are trained to please a human in the loop, so they excel at assistance and fail at unattended automation.
Ten month later and the doomsday scenario is playing full blast
I think AI as malware is a powerful way to describing it. May have uses in short term but carries the seeds of destruction with it.
What we call "data centres" now are very different from the data stores of several years ago - they are AI token mills and we really need to change the terminology. They are not passive repositories but vast factories generating responses to prompts. Language matters!
Meanwhile, academics across the world are thinking "This VC/President could have been replaced by a chatbot."
Wikimedia’s finds that OpenAI rogue agents paid a visit Wikimedia says agents it believes were operated by OpenAI made unapproved edits to its wikis, tried to turn its citation tool and Etherpad into proxies for fetching outside data, and sent millions of automated requests.... itnerd.blog/2026/10/07/wikimed…
Someone should start studying the use of AI-generated images for NIMBY causes. It’s honestly fascinating. It also shows most of these folks are clueless- the guy in the green shirt has three arms, for example
So I did a round of “AI annotator” freelance work. It’s a pretty bleak thing to put your time into. It’s the perfect storm of incompetence with people management, technical incompetence, ridiculously specific and thankless work, and hordes of people clamoring for said work. Getting a few bucks for my trouble is fine, but I would hate myself if I was forced to do this.
Billions in AI investment do not automatically equal regional development. The real test is whether communities gain lasting skills, infrastructure and power over their own future, Komla Tsey writes. #AI #RegionalAustralia
$100/mo AI agents, unavailable in the UK at launch, have a free open alternative. OpenDots — CopilotKit's MIT-licensed template — is a repo you clone, configure and run: no seat licence, no vendor pricing, bring your own model. The standout in my test was honesty: it returned clearly-marked blanks instead of inventing answers. A 28-cover restaurant got a reservations policy that knew which numbers were the owner's. #OpenSource #SelfHosted youtube.com/watch?v=jFnIKULEEoo
I think TESCREAL is going too far. It implies everyone in Silicon Valley is sitting there gleefully thinking about their imminently becoming a cyborg or a mind upload. I've also seen transhumanists who don't think a so-called superintelligence is imminent.
I gave Codex explicit instructions to use WxPython and native Windows controls. Last night I suddenly encountered an edit field that couldn't be navigated using braille. After working through several tests, I discovered Codex had ignored my instructions and used something called QT. I have submitted a refund request for the hours of time wasted testing controls that violated my explicit instructions. In my request, I pasted Codex's admition that it ignored my instructions and that I should get a refund. I also said I needed to do all of this testing because Codex isn't sufficiently trained on accessibility. While I wait to see if my refund request is addressed, I need to have days worth of work redone. Then, I will have to redo a lot of the speech and braille testing I thought had been finished. AI is a great tool. Sometimes, it's a huge pain in the ass.
I think there are still mathematical areas AI can’t help with. Things where we need totally new frameworks to conceptualize something, that do not currently exist. And I do think neuro needs those frameworks.
«We should see this wild and ecstatic boosterism — that AI is our salvation and destiny — as just the flip side of the doomerism that predicts our near-term extinction. They both flow downstream from a series of assumptions and beliefs, which when you see them close up, are genuinely weird, usually unsupported by actual evidence and often based on ethical assumptions and values that are simply bad.» talkingpointsmemo.com/edblog/w…
I shared this post by @pwnallthethings.bsky.social since the thread gives a different explanation for OpenAI's motivations for producing these maths results than I (and probably most of my colleagues in mathematics had been assuming). bsky.app/profile/pwna...
I always follow my music teacher’s rule: sit in the front row. It forces me to be engaged, pay attention and not drift. Yes, there was always a small risk of a tedious talk. But there is now an epidemic of people telling you what they asked ChatGPT to do, and you realize that you are poorly positioned to sprint for the exit.
I keep seeing posts saying, "Well, I don't know how to write a book, so I just use AI," or, "I can't draw backgrounds, and I don't have time, so I use AI." If you don't spend time *learning* how to do these things, you *never will do them.* How can you expect to be an artist if you never do art?
Folks asked how I keep a free AI app from bankrupting me. 😅 Guardrails: per-user + per-IP hourly limits, a daily circuit breaker, and a hard daily spend cap at OpenRouter itself. The bigger saving was switching models. Last 7 days, same prompts: Gemini 3.1 Flash Lite: $3.81 / 478 generations GPT-6 Luna: $0.39 / 467 generations 10x cheaper, better results. It's DX7 AI Studio: describe a sound, an LLM edits a real DX7 patch 🎹 dx7-patch-creator.pages.dev/ #DX7 #synth #LLM
My new favorite term: “wet-brained intentionality” (Alicia Pederson)— as opposed to dry-brained AI intentionality..
